The thing that finally broke me was a Renovate PR that bumped a single Helm chart version and produced eleven Terraform plan comments on a GitHub PR. Each one a wall of +, -, and ~ lines wider than a 14" laptop. Eleven, because we have 37 Terraform projects and 11 of them transitively depend on that chart. Technically the plan was correct. Practically, nobody on the team was going to read that. ...
In my previous post on Kubernetes metrics I walked through what the various metric endpoints in a cluster actually expose — kubelet, cAdvisor, kube-state-metrics, and friends. That post answered the what. This one answers the how: how do you actually collect all of it reliably, ship it to Grafana Cloud, and not blow your observability budget in the process? The answer we landed on is a self-contained Helm chart built around Grafana Alloy as the central scraping agent. Here’s how it works. ...